Finite distributive concept algebras

DOI 10.1007/s11083-006-9045-x

Finite Distributive Concept Algebras

Bernhard Ganter·Léonard Kwuida

Received: 23 August 2005 / Accepted: 5 October 2006 / Published online: 7 November 2006

© Springer Science + Business Media B.V. 2006

Abstract Concept algebras are concept lattices enriched by a weak negation and a weak opposition. In Ganter and Kwuida (Contrib. Gen. Algebra, 14:63–72,2004) we gave a contextual description of the lattice of weak negations on a finite lattice. In this contribution¹we use this description to give a characterization of finite distributive concept algebras.

Key words formal concept analysis·negation·concept algebras· weakly dicomplemented lattices·superalgebraic lattices

Mathematics Subject Classifications (2000) 03G25·03G10·06B15

1 Motivation

Weakly dicomplemented lattices have been introduced to capture the equational theory of concept algebras. These are bounded lattices equipped with two unary operationsand called weak complementation and dual weak complementation, and satisfying for all x and y the following equations:

(1) x≤x, (1’) x≥x,

(2) x≤y =⇒ x≥y, (2’) x≤y =⇒ x≥y, (3) (x∧y)∨(x∧y)=x, (3’) (x∨y)∧(x∨y)=x.

1Major parts are taken from [5], my PhD thesis. The reader is referred to [1] or [4] for an introduction to concept lattices.

Dedicated to I. Rival.

B. Ganter

Institut für Algebra, TU Dresden, D-01062 Dresden, Germany e-mail: Bernhard.Ganter@tu-dresden.de

L. Kwuida (

B

Mathematisches Institut, Universität Bern, CH-3012 Bern, Switzerland e-mail: kwuida@math.unibe.ch

The pair(x,x)is called the weak dicomplement of x and the pair(,)a weak dicomplementation. Concept algebras arose from the need to formalize the notion of “negation of a concept.” These are concept lattices equipped with two unary operationsandcalled weak negation and weak opposition, and defined for each formal concept(A,B)by

(A,B):=(A,A) and (A,B):=(B,B),

where X stands for the set-complement of X andthe derivation in a formal context (see [6] for further details). Recall that a formal context is a triple(G,M,I)of sets such that I⊆G×M. The members of G are called objects and those of M attributes.

If(g,m)∈I the object g is said to have m as an attribute. For subsets A⊆G and B⊆M, Aand Bare defined by

A:= {m∈M| ∀g∈A gIm} and B:= {g∈G| ∀m∈B gIm}.

A formal concept of the context(G,M,I)is a pair(A,B)with A⊆G and B⊆M such that A=B and B=A. In this case A is called the extent and B the intent of the concept(A,B). The set of all formal concepts of the context(G,M,I)is denoted by B(G,M,I).γg:=({g},{g})andµm:=({m},{m})denote special concepts called respectively object concept and attribute concept. The hierarchy on concepts is captured by the inclusion relation on extents. In fact a concept (A,B) is called a subconcept of a concept(C,D)provided that A⊆C (which is equivalent to D⊆B).

We also call(C,D)a superconcept of(A,B)and write(A,B)≤(C,D). The poset (B(G,M,I); ≤)is a complete lattice and is called the concept lattice of the context (G,M,I)(see [4]). The concept algebra of a contextKwill be denoted byA(K). i.e.

A(K):=(B(K),∧,∨,,,0,1). The weak operations onA(K)satisfy the equations (1)to(3) above (see [6]). Concept algebras motivated the introduction of weakly dicomplemented lattices. Other examples are Boolean algebras (by duplicating the complementation), distributive double p-algebras, bounded lattices (x=1for x= 1 and x=0for x=0, usually called the trivial weak dicomplementation), etc. . . . Thus the same bounded lattice might carry many weak dicomplementations.

On a lattice L a weak dicomplementation(¹,¹)is said to be finer than a weak dicomplementation(²,²), written(¹,¹)(²,²), if x¹≤x²and x¹≥x²for all x in L. The “finer than” relation is an order relation on the set Wd(L) of all weak dicomplementations on a bounded lattice L. The poset(Wd(L),)admits a top element (the trivial weak dicomplementation). If the set J(L)of all completely join-irreducible elements is -dense and the set M(L) of all completely meet- irreducible elements is

-dense²then the weak dicomplementation of the concept algebraA(J(L),M(L),≤)(often mentioned as standard weak dicomplementation) is the bottom element of the poset(Wd(L),). In this case the poset of weak dicom- plementations is a complete lattice. In fact for a nonempty family

(ⁱ,ⁱ)|i∈I of

2This is always the case for finite lattices. We often refer to this property as “lattices with enough irreducible elements.” A special subclass is that of doubly founded lattices.

weak dicomplementations on such a lattice L, the supremum is the operation(^I,^I) defined for all x∈L by

x^I := xⁱ |i∈I

and x^I:= xⁱ |i∈I .

Analogously the set Wc(L) of weak complementations on such an L forms a complete lattice and contains the class of weak negations as a complete sublattice [5, Theorem 3.2.1]. Weak negations and weak oppositions behave independently (see Lemma 1 below). Thus the lattice of representable weak dicomplementations is a product of the lattice of weak negations and the lattice of weak oppositions. Since we are interested in representable weak dicomplementations, we can just concentrate on weak complementations.

Lemma 1 [3, Lemma 3] Let L be a complete lattice andK:=(G,M,≤)be a subcon- text of(L,L,≤)such thatB(K)∼=L. For g and m in L we set

Kg:=(G∪ {g},M,≤)andK^m:=(G,M∪ {m},≤).

We have ^Km =^Kand^Km ≤^K, as well as^Kg =^Kand^Kg ≥^K.

If a weakly dicomplemented lattice is (isomorphic to) a concept algebra of some context it is said to be representable³ (by this context). We will also speak about representable weak complementations. However not all complete lattices satisfying the equations (1) to (3) are (isomorphic to) concept algebras. In this contribution we prove that finite distributive concept algebras are charaterized by the aforementioned equations. A start up example is the lattice product of a two-element and a n-element chain. Here all weak complementations are completely determined by the image of a single element, and are all weak negations (see Section2). This section addresses the problem of determining weak dicomplementations. Section3is devoted to a contextual representation of the lattice of weak negations. Afterwards we examine in Section 4 the influence of the underlying lattice on the sublattice of representable weak complementations and then establish the characterization of finite distributive concept algebras.

2 Determination of Weak Dicomplementations

We start with a simple structure, a product of a two-element and an n-element chain.

2.1 Weak Dicomplementations on the Lattice 2×n

We want to determine all weak dicomplementations on the lattice L:=2×n, the product of a two-element chain and an n-element chain. We use the labeling on

3In this contribution we use the term “representable” in the sense of the “strong representation problem” for weakly dicomplemented lattices (see [5, Section 1.4]).

Figure 1 Weak

dicomplementations on 2×n.

Figure1. L is a distributive lattice. The axioms of a weak complementation f can be rewritten as follows:

(1) f²x≤x,

(2) x≤y =⇒ f x≥ f y and (3) x∨ f x=1.

Therefore x≤t implies f x=1 and r≤x≤s implies f x≥u, as well as f u≥r.

We set A:= {x∈ [r,s] | f x=1}. If A is empty then f x=u for all x∈ [r,s]. Since f²r≤r holds we obtain the equality f u=r. We assume that A is not empty and denote by x1the greatest element of A. For all x in[r,s], we have

x≤x1 =⇒ f x=1and x>x1 =⇒ f x=u.

We denote by x2 the successor of x1. Since f²x2≤x2 we obtain f u≤x2; if f u<

x2 we would have f²u=1>u, a contradiction. Then f u=x2. The operation f is completely determined by the image of u in [r,1]. There are exactly n such operations (which are all weak complementations). Similarly there are also n dual weak complementations g on L, each determined by the image of r in [0,u].

Thus Wd(L)is the product of two n element chains (and contains exactly n²weak dicomplementations). They are all representable weak dicomplementations. In fact for a weak complementation f and a dual weak complementation g on L the concept algebra of the context(J(L)∪ {f u},M(L)∪ {gr},≤)is (isomorphic to) the weakly dicomplemented lattice(L,∧,∨,f,g,0,1)since

u≡

((J(L)∪ {f u})\ ↓u), ((J(L)∪ {f u})\ ↓u)

=(↓f u∩(J(L)∪ {f u}),↑f u∩(M(L)∪ {gr}))

≡ f u,

where the relation ≡ identifies a lattice element with its corresponding formal concept. Dually r≡gr.

2.2 Determining Weak Dicomplementations

We want to characterize weak dicomplementations on a given complete lattice L.

Since clarifying⁴ a context does not alter the concept algebra structure, a repre- sentable weak dicomplementation on L can always be represented by a pair(G,M) of subsets of L such that the concept algebra of the context(G,M,≤)is isomorphic to the given representable weakly dicomplemented lattice. To avoid confusion we sometimes index representable weak operations by their context name. This is usually the case if we deal with more than one context. A weak complementation on L is represented by a subset G of L iff G is supremum dense and for all x in L, x=

{g∈G|g≤x}. Dually, a dual weak complementationon L is represented by M⊆L iff M is infimum dense and for all x in L, x=

{m∈M|m≥x}. The problem of finding a characterization of representable weak dicomplementations is still open. The idea is to first determine all weak complementations on L, and then check whether the lattice of weak complementations, via this determination, can be mapped to its sublattice of weak negations. The upcoming proposition gives an insight on which subsets can represent which weak dicomplementations. For a weak complementation, an element u∈L is said to be-compatible if u≤x or u≤x for all x∈L. Note that all∨-irreducible elements are-compatible.

Proposition 1 If G representson L then G∪H also representsif and only if all elements of H are-compatible.

Proof x=

{g∈G|g≤x} ≤

{u∈G∪H|u≤x}. The inequality is proper iff there is some u∈H with u≤x and u≤x. Thus u must be -incompatible.

Conversely if u∈H is incompatible, then u≤x and u≤xfor some x, and for this

x the inequality is proper.

Proposition 1 shows that the set of-compatible elements is the best candidate set to represent. A necessary condition is to be

-dense in L.

In general a weak dicomplementation on L can be determined by its values on some subsets of L.

Lemma 2 [3] Letbe a weak complementation on L and M a

-dense subset of L.

For all u∈L we have (a) m|m≥u,m∈M

=u= n|n≥mfor all m∈M,m≥u . (b) Weak complementations are determined by their values on any

-dense subset.

(c) Weak complementations are determined by their ϒ-relation (on any -dense subset) defined by mϒn: ⇐⇒ n≥m.

For a context(G,M,I)we define the relation⊥on M by m⊥n: ⇐⇒ m∪n=G.

4A context is clarified if for g and h (both objects or both attributes) g=h holds only if g=h holds. An object (resp. attribute) g is reducible if there is a set X of objects (resp. attributes) such that g=Xholds. A clarified context is reduced if all its objects and attributes are irreducible.

Lemma 3 Let L be a complete weakly complemented lattice such that a subset G of its set of-compatible elements is

-dense. Let M be a

-dense subset of L. Then B(G,M,≤)is isomorphic to L and theϒ-relation ofis an order filter⁵of the⊥- relation of(G,M,≤).

Proof Let be a weak complementation on L and G a subset of the set of - compatible elements of L that is

-dense. Let M be a

-dense subset of L; the concept latticeB(G,M,≤)is isomorphic to L and for all x∈L we have,

x^(G,M,≤)=

{g∈G|gx} ≤x. (†) Let(m,n)∈ϒ. From mϒn we get

m≥n≥n^(G,M,≤) by(†).

For any element g∈G, if g∈n= {h∈G|h≤n}then gn. Thus g≤n≤m and g∈m. Thus mϒn implies m∪n=G, and so m⊥n.

i.e.(m,n)∈ϒ =⇒ (m,n)∈ ⊥.

Now assume that (m,n)∈ϒ and (x,y)≥(m,n). We have m≥x and y≥n≥ m≥x; thus y≥x, and xϒy. This shows thatϒis an order filter of⊥. Does the converse hold, i.e., do theϒ-relations exhaust all order filters of⊥? A positive answer would imply the distributivity of the lattice of weak complementa- tions (see Birkhoff’s theorem Section4).

Remark 1 In particular, if L is a finite lattice, then its set of∨-irreducible elements is a

-dense subset of-compatible elements for any weak complementation on L.

It is minimal. Theϒ-relation of any weak complementation on L is an order filter of the⊥-relation of the reduced context of L. Having a smallest

-dense subset gives the opportunity to have allϒ-relations as order filter of the same⊥-relation on any fixed

-dense subset of L. Note that the smaller the set of objects is, the larger the

⊥-relation is.

The relationsϒand⊥are symmetric. In the rest of this contribution we adopt the following notations:

:= {{m,n} ⊆M|mϒn} and T:= {{m,n} ⊆M|m⊥n}.

5Theϒ-relation and the⊥-relation are both defined on M, and are subsets of M×M (ordered componentwise).

Remark 2 If we consider the pair inT to be of distinct elements, then T can be empty.

In this case, T= ∅iffis the trivial weak complementation. In fact, T= ∅ ⇐⇒ for all m∈M, m∪n=G for all n∈M\ {m}

⇐⇒ for all m∈M, G\mnfor all n∈M\ {m}

⇐⇒ for all m∈M, (G\m)nfor all n∈M\ {m}

⇐⇒ for all m∈M, mn for all n∈M\ {m}

⇐⇒ for all m∈M, m=1.

In the rest of this paper we assume that T is nonempty.

3 Lattice of Representable Weak Complementations

We denote by Ext(K)the lattice of all extents of the contextK:=(G,M,I). The contextR:=(Ext(K),T,R)is defined by:

UR{m,n} : ⇐⇒ U ⊆mor U⊆n.

(−)^Rdenotes the derivation operation inR. We set(−)^c:=(−)^RR. For U in Ext(K) we have

{U}^c= {W∈Ext(K)|UR{m,n} =⇒ WR{m,n}for all{m,n} ∈T}.

The relation I ofKis extended by Ieon(GExt(K),M,Ie)as follows:

Ie∩G×M=I and U Iem: ⇐⇒ U ⊆m for U∈Ext(K).

Iewill also denote its restriction on(GH,M,Ie)withH⊆Ext(K). These contexts have isomorphic concept lattices and define the same weak opposition. However the weak negations they define can be different (see Lemma 1). As above (−)^Ie denotes the derivation in the extended contexts. Recall that an extent U is called

-compatible if and only if U⊆A or U⊆ Afor all extents A, and that the weak negation of an extented context coincides with the old weak negation if and only if all new objects are-compatible in the old context. Here is a characterization of the compatibility by means of the⊥-relation.

Lemma 4 For U ∈Ext(K)

U∈T^R ⇐⇒ U is-compatible inA(K) ⇐⇒ ^(G,M,I)=(G{U},M,Ie)

Proof The second equivalence follows from Proposiotion 1. The first equivalence is obtained by contraposition: suppose U∈/T^R; then there is{m,n} ∈T such that U mand Un; for A:=mwe have U A and U n⊇A. Conversely suppose that U is-incompatible inB(K); then there is A∈Ext(K) such that UA and UA. There exist m∈Awith Um, and n∈ Awith Unsuch that m∪n⊇

A∪A=G.

Lemma 5 For H⊆Ext(K), H^c is the set of ^(GH,M,Ie)-compatible elements. Hence,

(GH,M,Ie) =^(GHc,M,Ie).

Proof

W ∈H^c ⇐⇒ ({m,n} ∈T and UR{m,n}for all U ∈H =⇒ WR{m,n})

⇐⇒ (m∪n=G and(U⊆mor U ⊆n)∀U∈H =⇒ W ⊆m or W⊆n)

⇐⇒ (m^Ie∪n^Ie=GH =⇒ W⊆mor W⊆n)

⇐⇒ W is^(GH,M,Ie)-compatible.

Lemma 6 Let L be a lattice with two weak complementations¹and². If²≤¹then every¹-compatible element is also²-compatible.

Proof Let u be a¹-compatible element. For all x∈L, u≤x or u≤x¹. This implies u≤x or u≤x²since x¹≤x². Thus u is²-compatible.

From this, we can now prove the following result

Theorem 1 [Contextual description of the lattice of weak negations] Let Kbe a reduced⁶ context. The concept lattice of the context(Ext(K),T,R)is isomorphic to the lattice of representable weak complementations onB(K).

Proof We denote by Wn(L)the set of representable weak complementations on a lattice L. By Lemma 1 and Lemma 5 the assignment

ψ:(H,H^R)→^(GH,M,Ie)

defines an increasing map fromB(Ext(K),T,R)to Wn(B(K)). LetH1andH2be subsets of Ext(K). We assume that

(GH1,M,Ie)=^(GH2,M,Ie); then ¹:=^(GHc1,M,Ie)=^(GHc2,M,Ie)=:².

As¹=²we have (by Lemma 5) for each W∈Ext(K),

W ∈H^c1 ⇐⇒ W is¹-compatible ⇐⇒ W is²-compatible ⇐⇒ W ∈H^c2. Thus the map

φ:^(GH,M,Ie)→(H^c,H^R)

is well defined from Wn(B(K))toB(Ext(K),T,R). The compositions ψ◦φ and φ◦ψare identity maps. In fact,

φ◦ψ(H,H^R)=φ(^(GH,M,Ie))=(H^c,H^R)=(H,H^R) for all(H,H^R)∈B(R),

6See Remark 1.

Figure 2 Free distributive lattice generated by three elements and the corresponding reduced context.

and for allH⊆Ext(K)we have

ψ◦φ(^(GH,M,Ie))=ψ(H^c,H^R)=^(GHc,M,Ie)=^(GH,M,Ie) by Lemma 5.

Thusψandφare bijections and inverse to each other. To achieve the proof it remains to show thatφis also increasing. To see this assume that

1:=^(GH1,M,Ie)≤ ^(GH2,M,Ie)=:².

By Lemma 6 the set H^c2 of ²-compatible extents contains the set of ¹-compatible extents H^c₁. Therefore^(GH1,M,Ie) ≤ ^(GH2,M,Ie)impliesH₁^c⊆H^c₂andφis increasing.

Thusφandψare order preserving bijections, inverse each other and are then lattice

isomorphisms.

Before we proceed further let us have a look at an example.

Example 1 We consider the free distributive lattice generated by three elements.

Its reduced context has the attribute set M:= {7,8,10,14,15,16}and the object set J:= {1,2,3,7,8,10}(see Figure2).

The set of orthogonal pairs of attributes is given by

T= {{14,15},{14,16},{14,10},{15,16},{15,8},{16,7}}.

The contextRis given on Figure3. All objects from0to10, and the object17are reducible. The resulting context is a copy of the context on Figure2. Thus the lattice of concrete weak complementations on the free lattice generated by three elements is isomorphic to this lattice.

In the next section we show that some properties of the initial lattice can be carried over to the lattice of representable weak complementations.

4 Distributive Lattices

We first recall some results needed in this section. Completely distributive complete lattices in which the set of (completely)

-irreducible elements is

-dense are called superalgebraic, being characterized by the property that every element is a join of supercompact (i.e.

-prime) elements. J(L)denotes the set of supercompact elements of L.

Birkhoff’s theorem. If a lattice L is superalgebraic then x→ ↓x∩J(L) des- cribes an isomorphism of L onto the closure system of all order ideals of (J(L),≤). Conversely for every ordered set (P,≤) the closure system of all order ideals of (P,≤) is superalgebraic.

Contraordinal scale. For an arbitrary ordered set (P,≤) the context (P,P,) (called contraordinal scale) is reduced. Its concept lattice is a copy of the lattice of order ideals of(P,≤).

Contranominal scale. For a set S the context (S,S,=) (contranominal scale) is reduced. Its concept lattice is isomorphic to the power set lattice of S.

Let L be a superalgebraic lattice. There is a poset (P,≤) such thatB(P,P,)is isomorphic to L. The relation⊥on P is characterized by

m⊥n ⇐⇒ ↑m∪ ↑n=P ⇐⇒ ↑m∩ ↑n= ∅.

Lemma 7 The relation≤defined on T= {{m,n} ⊆ P|m⊥n}by {x,y} ≤ {s,t} : ⇐⇒ {x,y} ⊆ ↓{s,t}

Figure 3 Context of all weak negations on the free distributive lattice generated by three elements.

is an order relation.

Proof Reflexivity and transitivity are obvious. To prove antisymmetry, we assume {x,y} ≤ {s,t}and{s,t} ≤ {x,y}. Note that x and y cannot together be less than s or than t; otherwise s or t would belong to↑x∩ ↑y which is empty. Even the assertion

“x≤s, y≤t and s≤y, t≤x” cannot hold; otherwise we would have x≤s≤y≤ t≤x which is a contradiction. Without loss of generality our assumption implies x≤ s,y≤t and s≤x,t≤y; therefore{x,y} = {s,t}and ≤is antisymmetric. Thus ≤is

an order relation on T.

Lemma 8 For any poset (P,≤) the intents of (Ext(P,P,),T,R) are exactly the order filters of(T,≤).

Proof Let U ∈Ext(P,P,) with {m,n} ∈U^R and {x,y} ≥ {m,n}. On one hand {m,n} ∈U^Rif and only if U ⊆mor U⊆n; up to permuting m and n we have

{x,y} ≥ {m,n} ⇐⇒ x≥m and y≥n ⇐⇒ ↑x⊇ ↑m and↑y⊇ ↑n. Hence,{x,y} ≥ {m,n}implies x⊇m⊇U or y⊇n⊇U , and by then{x,y} ∈U^R. Thus U^R is an order filter. For any intent B of (Ext(P,P,),T,R), there isU ⊆ Ext(P,P,) such that B=U^R= {U^R|U∈U}, and is an order filter. Thus all intents of(Ext(P,P,),T,R)are order filter of(T,≤).

Conversely, let B be an order filter of(T,≤). We want to prove that B=B^RR. Let{m,n}∈/B. We have

{m,n}∈/B ⇐⇒ ∀{s,t} ∈ B,{s,t}{m,n}

⇐⇒ ∀{s,t} ∈ B,ms and ns or mt and nt

⇐⇒ ∀{s,t} ∈ B,↓{m,n} ⊆sand↓{m,n} ⊆t

⇐⇒ ∀{s,t} ∈ B,↓{m,n}R{s,t}

⇐⇒ ↓{m,n} ∈ B^R.

Moreover, ↓{m,n}m and ↓{m,n}n imply not(↓{m,n}R{m,n}). Thus

{m,n}∈/ B^RR. Hence, B=B^RR.

We denote by oF(T,≤)(resp. oI(T,≤)) the superalgebraic lattice of order filters (resp. ideals) of(T,≤), and write∼=^dto mean “is dual isomorphic to”. Int(K)denotes the lattice of intents of a contextK. Note that Ext(K)∼=^dInt(K).

Theorem 2 The lattice of representable weak complementations on any superalge- braic lattice is superalgebraic.

Proof Let L be a superalgebraic lattice. There is a poset(P,≤)(of supercompact elements of L) such that B(P,P,)∼=L. By Theorem 1 the lattice of repre- sentable weak complementations on L is isomorphic to B(Ext(P,P,),T,R).

Using Lemma 8 we get

B(Ext(P,P,),T,R)∼=^dInt(Ext(P,P,),T,R)=oF(T,≤)∼=^doI(T,≤),

which is superalgebraic (by Birkhoff’s theorem).

Corollary 1 The lattice of representable weak complementations on a complete atomic Boolean algebra (resp. a finite distributive lattice) is a complete atomic Boolean algebra (resp. a finite distributive lattice).

Proof Note that finite distributive lattices and complete atomic Boolean algebras are special cases of superalgebraic lattices. In the second case, the context(P,P,) is exactly(P,P,=), and the lattice of representable weak complementations is a copy of the lattice of order filter of the poset(T,=), which is a complete atomic Boolean

algebra.

We are now ready to give the main results of this contribution.

Theorem 3 For any superalgebraic lattice L, the weak complementations on L form a superalgebraic lattice isomorphic to the lattice of all order ideals of

T= {{m,n} ⊆P| ↑m∩ ↑n= ∅}, where P is the set of all

-prime elements of L and T is ordered by {x,y} ≤ {m,n} ⇐⇒ {x,y} ⊆ ↓{m,n}.

Proof Letbe a weak complementation on a superalgebraic lattice L. By Lemma 2,

is completely determined by itsϒ-relation on P. Thisϒ-relation is, by Lemma 3, an order filter of the⊥-relation on P. Theϒ-relation and⊥-relation are symmetric and are exactly determined by their respective factorization (with respect to the symmetry)

= {{m,n} | m,n∈M,mϒn} and T= {{m,n} | m,n∈M, m⊥n}.

The factorisation of the order≤on⊥corresponds to the order≤on T (see Lemma 7) and turnsinto an order filter of the poset(T,≤). The assignmentη:→η():=

T\defines a map from Wc(L)to oI(T,≤). By Lemma 2 we have

1≤²⇐⇒ ϒ1⊇ϒ2 ⇐⇒ 1⊇2 ⇐⇒ T\1⊆T\2 ⇐⇒ η(¹)≤η(²).

Hence,ηis an order embedding of Wc(L)into oI(T,≤).

Conversely if J is an order ideal of(T,≤), then

T\J∈oF(T,≤)=Int(Ext(P,P,),T,R) by Lemma 8.

Thus

(T\J)^R, (T\J)

∈B(R)andψ

(T\J)^R, (T\J)

∈Wn(L)⊆Wc(L). Thus ζ: J→ψ

(T\J)^R, (T\J)

defines a map from oI(T,≤)to Wc(L). Moreover,η◦ ζ(J)=J. In fact, settingH=(T\J)^Rwe have

ζ(J)=ψ(H,H^R)=^(PH,P,e) and η◦ζ(J)=T\ζ(J)=T\_(PH,P,e).

Thus η◦ζ(J)=J iff T\(PH,P,e) =J iff (PH,P,e) =T\J. Thus we should prove that{m,n}∈/ J iffµm≥µn^(PH,P,e)for all{m,n} ∈T. Note that

{m,n}∈/ J ⇐⇒ {m,n}{a,b} ∀{a,b} ∈ J ⇐⇒ {m,n} ∈ J.

Thus T\J=JandH= J^R.

_(PH,P,e) = {{m,n} ∈T|µm≥µn^(PH,P,e)}

= {{m,n} ∈T|(m^e,m^ee)≥(n^e,n^ee)^(PH,P,e)}

=

{m,n} ∈T|(m^e,m^ee)≥

(PH\n^e)^ee, (PH\n^e)^e

=

{m,n} ∈T|m^e ⊇(PH\n^e)^ee

=

{m,n} ∈T|m^e ⊇PH\n^e

=

{m,n} ∈T|x∈/n^e implies x∈m^efor all x∈PH

=

{m,n} ∈T|not(xen)implies xem for all x∈PH

= {{m,n} ∈T|(not(xn)implies xm for all x∈P)and (not(Uen)implies Uem for all U∈H)}

= {{m,n} ∈T|(not(x∈n)implies x∈mfor all x∈ P)and (not(U⊆n)implies U⊆mfor all U ∈H)}

= {{m,n} ∈T|(x∈/nimplies x∈mfor all x∈P)and (U ⊆nor U⊆mfor all U ∈H)}

= {{m,n} ∈T|UR{m,n}for all U ∈H}, since m∪n=P

= {{m,n} ∈T|UR{m,n}for all U ∈J^R}

= J^RR=(T\J)^RR=T\J, since T\J∈Int(R).

Thusηis an isomorphism of Wc(L)onto oI(T,≤).

Theorem 4

(a) The lattice Wc(L) of weak complementations on a superalgebraic lattice L is isomorphic to its sublattice Wn(L)of representable weak complementations.

(b) On finite distributive lattices all weak complementations are representable. i.e.

each weak complementation on a finite distributive lattice is a weak negation.

(c) All finite distributive weakly dicomplemented lattices are (isomorphic to) concept algebras.

Proof (a) follows from

Wc(L)∼=oI(T,≤)∼=^doF(T,≤)=Int(R)∼=^dB(R)∼=Wn(L), whereRis the context(Ext(P,P,),T,R)and Int(R)the lattice of its intents.

In the finite case Wc(L) is finite and has the same cardinality as its subset Wn(L). Thus Wn(L)=Wc(L)and (b) is proved. i.e. All weak complementations are weak negations. Dually we obtain that each dual weak complementation is a weak opposition and (c) is proved (see Lemma 1 and Proposition 1).

5 Conclusion

The main result of the present contribution states that finite distributive concept algebras are exactly finite distributive lattices with two unary operations and satisfying the following equations for all x and y:

(1) x≤x,

(2) x≤y =⇒ x≥y, (3) (x∧y)∨(x∧y)=x,

(1’) x≥x,

(2’) x≤y =⇒ x≥y, (3’) (x∨y)∧(x∨y)=x.

In this case the condition(3)is equivalent to y∨y=1and the condition(3)is equivalent to y∧y=0. Now that we have a characterization of finite distributive concept algebras the next step would be to consider standard problems such as free structures, decomposition, etc. . . . Congruences of finite distributive concept algebras have been described [2]. Although Wn(L) and Wc(L)are isomorphic for a superalgebraic lattice L, it is still not clear whether they are equal.

References

1. Davey, B.A., Priestley, H.A.: Introduction to Lattices and Order. Cambridge University Press, New York, Second Edition (2002)

2. Ganter, B.: Congruence of finite distributive concept algebras. In: Lecture Notes in Computer Science, vol. 2961, pp. 128–141. Springer, Berlin Heidelberg New York (2004)

3. Ganter, B., Kwuida, L.: Representable weak dicomplementations on finite lattices. Contrib. Gen.

Algebra 14, 63–72 (2004) (J. Heyn Klagenfurt)

4. Ganter, B., Wille, R.: Formal Concept Analysis: Mathematical Foundations. Springer, Berlin Heidelberg New York (1999)

5. Kwuida, L.: Dicomplemented lattices. A contextual generalization of Boolean algebras. Shaker Verlag, Aachen (2004)

6. Wille, R.: Boolean concept logic. In: Lecture Notes in Computer Scince, vol. 1867, pp. 317–331.

Springer, Berlin Heidelberg New York (2000)